When scaling the first bottlenecks to run into are
-
CPU usage: Python based relay and cache use lots of CPU for parsing
- Solve by switching to compiled versions for example
-
CPU usage: When running relays and caches on the same host
- Decouple on different hosts
When you scale beyond Graphite
- InfluxDB can ingest Graphite Traffic
- M3DB allows you to continue to use Graphite Queries against its TSDB
- Replacing Graphite with Prometheus